PrefaceOne fine spring afternoon in 1982, I was sitting in John McDowellsliving room at 17 Merton Street in Oxford, explaining to him why I hadnever found the private language argument very convincing. A fewpatient and searching questions later, it suddenly occurred to me first,that perhaps there was a good argument there after all; second, that itwould be an important fact if there were, since the argument points inthe right direction on some central topics in philosophy where thetemptation to make bad starts has proved powerful in recent times—and not only among philosophers,- and third, that I should write a bookabout this. I believe that the first: two of these hunches have proved tobe correct. Part of the importance of an argument against private language is inpointing away from a certain conception of our psychologies, andparticularly of our attitudes—those usually referred to as preposi-tional and others. On the conception, we fix the facts about, say, ourbeliefs, intentions and expectations—and particularly their content orsemantics—by ourselves, and independent of the perspectives othersmight have on us; as it were, simply in virtue of the way we areconstituted on the inside, or (perhaps even literally) intradermally—together, if you like, with the internal history of that constitution. Onsome versions of this conception, the emphasis is placed on the pointthat if, for example, Odile believes that Montreal is east of Cincinnati,then for her to do so is for her to be in some particular intradermalstate, specifiable (individuable) in other terms—neurologically, per-haps, or by a machine table, or something of the sort. As it issometimes put, her believing what she does must consist in herstanding in some specified relation to a syntactically specified object,thus one which is a fit subject for computations—as beliefs surelyare—and moreover to formally specified ones. Whether Odiles beliefhas one semantics or another would then be derivable (modulo causalhistories) from the nature of the otherwise -specified state involved.On other versions, the point of emphasis is that the content of Odilesbelief (at least where true exact determination of that content is atstake) is determined by some domain of inner facts which are open to
viii Prefaceobservation—or anyway direct observation—only by her; as it were,by the precise course of her experience, given the way that experiencefeels from the inside. It is a short step from this to the view thatseeing, or knowing what the precise content of her belief is, demandsstanding in a relation to some special domain of inner facts in whichonly Odile herself can stand. So to know, or perhaps even to be able toappreciate precisely what it is that Odile believes, one would have todo nothing less than be Odile herself. (At this point, the connectionwith private language is already obvious.) As McDowell has recentlyargued1—in a way that is both incisive and illuminating—these twoversions are really but slightly different elaborations of the sameunderlying view. What the present book will argue is that thatunderlying conception involves an essentially false view of howpropositional attitudes come by their semantics; that it is essential towhat these attitudes are that their content be assigned from theoutside by reacting to facts that are available equally to all of us. A second central import of the private language argument is for acertain view of the content or semantics (or, if you insist, meanings) ofpublic words—yours and mine, say. On that view, the content ofOdiles words, Montreal is east of Cincinnati—where she informsHugo of this, say—is uniquely fixed (again, if you like, modulo causalhistories) by her attitudes towards them—notably, on most views, bysome relevant set of her intentions and beliefs. This sort of view hasbeen popularized by, and is rightly associated with the name of H. P.Grice. It takes more work than actually will be done in this book toshow that such views contain, strictly speaking, literal falsehoods. Butthe argument to come certainly suggests that they reverse thedirection of fit. For they at least suggest that in asking after the contentof words, we are free to appeal to semantic facts about attitudes of thespeaker, where these are available anyway, no matter what the solvingof further problems as to the disambiguation of words might show.Whereas, it will be argued here, there are exactly the same problems topose for the semantics of attitudes or thoughts as there are for thesemantics of words;- further, their solutions lie along exactly the samepaths, and via appeals of the same sort to just the same further facts.Moreover, the Gricean view suggests that once we do fix the right factsabout the speakers attitudes, and again modulo causal stories, therecan be only one semantics for the words in question which is thesemantics that those attitudes require, so that there is a way of justreading off the semantics of the words from the right set of attitudes(with their semantics fixed). One of the main theses in what is to come 1 See McDowell [1986).
Preface ixis that, in an important sense, that simply could not be. In that sense,the Gricean story has not identified the factors on which the semanticsof words depends. I have elaborated here on these matters of importance since, as it hasturned out—unfortunately—they will receive much less attention inthe main text than they merit. In fact, Gricean views have turned outnot to get discussed at all. Just getting to the point of being able todiscuss these questions properly has been enough of an exercise for thepresent. A fuller discussion of them will, I hope, be the subject of asequel. I have used the discussion of private language, in this book, as a focalpoint for a discussion of Wittgensteins opposition to a certain pictureof semantics—a picture shared by Frege, by Russell, and by Wittgen-steins earlier Tractarian self, among many others—and of the positivealternative to that picture that Wittgenstein proposes in the Investi-gations. Accordingly, I have read the argument as directed against thatpicture. (Note that the pictures in question are pictures of semantics:not of this or that sort of semantic item, but of semantic properties andwhat it is to have them in general.) Of course, semantics is not the onlytopic of the Investigations. Nor does the present topic exhaust theimport of the private language discussion. That discussion certainlylooks as if it is about certain topics in the philosophy of mind: picturesof sensations and the like,- and more generally pictures of psychologicalproperties. No doubt that is correct. In fact, there is a connectionbetween the semantic themes here and those pictures of our mentallives—a connection which is at least stated in the last section of thelast chapter of this book. But I think it no disgrace to admit that thereis much more in the Investigations than is touched on here; and Icertainly do admit it. There is certainly little enough about semantics on which Frege andRussell agreed. The differences between their views are useful forlocating that common property which otherwise is liable to gounnoticed. I will not say at this point what their shared picture ofsemantics is, nor what Wittgensteins alternative to it is. I have statedthe first point, and the core of the second, as clearly as I am now able toin chapter 1. There is no point in mere repetition. But the differencesbetween Frege and Russell also deserve some comment. For on the faceof them, they certainly suggest that any private language argumentwould apply to the one philosopher differently than it does to theother. In those differences, much of the further import of the argumentlies. If one must choose, I think it is probably better to see the maintarget of the current argument to be Frege; though Russell is far frombeing spared by it.
x Preface The application of a private language argument to Russell seemsobvious. For, after all, Russell insisted that language must be private,and that it would be absolutely fatal 2 if it were not. Language must beprivate, on Russells view, because at the bottom of the path of thelogical analysis he then believed in, reference must be to privateobjects. As we will see (in chapter 2), reference to private objects is notan essential feature of private language. But it is certainly a sufficientone. (On one currently popular view of Russell, it is possible to extractthis aspect of his views on reference, and delete it as an anachronisticembarrassment, while retaining a substantial and important part ofthose views intact. Since Russell had arguments purporting to showthat names, at least, must refer to private objects, it is not obvious thatthis is so. What is the response to these arguments? Later, particularlyin chapters 6 and 7, we will consider what sort of response to them isanyway available, and just how much of Russells view of names thensurvives, and on what conditions. The result will be, inter alia, asomewhat different view of current doctrines of direct reference.) If private language is impossible, then there must be somethingwrong with Russells views. It is worth noting that 1918—the year inwhich Russell propounded these views in a series of lectures inLondon3—is the same year in which Frege published an extendedattack on the idea of private language4. Freges official view, at least, isthus that our languages, at least, and more generally any language, fiteither for communication or for the expression of the thoughts whichconcern us (for example, thoughts about mathematics) must be public.Freges picture of semantics, whatever it may be, is surely meant to fitat least some such languages. So if one wanted to show that picture tobe wrong, it would not do simply to produce a private languageargument and stop. As far as the conclusion of the argument isconcerned, Frege would just read that as something which had beenpart of his view all along. It may well be that Frege was committed covertly to privatelanguage. There are, for one thing, his well-known views on the senseswe attach to the first person pronoun, and to proper names. But thesemight be regarded as peripheral and unfortunate appendages on hiscentral system. I do mean to argue in what follows that Freges picture 2 Russell , p. 195: When one person uses a word, he does not mean by it thesame thing as another person means by it. I have often heard it said that that is amisfortune. That is a mistake. It would be absolutely fatal if people meant the samething by their words, . . . because the meaning you attach to your words must depend onthe nature of the objects you are acquainted with, and . . . different people are acquaintedwith different objects. 3 Russell , 4 Frege ,
Preface xiof semantics is ultimately incompatible with the demands of pub-licity. But, in terminology suggested by Michael Hallett, this is not somuch because Frege thought in terms of one (or higher finite N)speaker languages, but rather because he thought, as it were, in termsof no-speaker languages. For Frege, any semantics for language orthought to have is the semantics it is, and requires what it does, inessential respects, at least, in the nature of the case, and quiteindependent of any reactions we might have to it. So the semantic factsof a language being what they are does not require there being anyspeakers of that language at all. A language may be just there to bespoken, semantically full-blown and insulated from any effects on itthat speaking it might have. Which no-speaker languages there are to bespoken—that is, which semantics there are for some language or otherto have—is, on the Fregean view, a fixed point. The question for anatural language such as English is: with which no-speaker languagedoes it share a semantics? It is at that level only that the reactions of itsspeakers may play a role. For all that, the semantic of a language, or thesenses of its expressions, is meant to be something available for all ofus to grasp. The domain of senses is intrinsically a public domain.What will be argued is that semantic facts could not be simultaneouslythus graspable and thus insulated from our reactions to them. Theimpossibility of private language then eliminates one of the alter-natives that leaves. In any event, the present strategy is not to argue that Frege was reallysome sort of closet private linguist. Rather, the function of privatelanguage in the present discussion is quite different. The first task inwhat follows is simply to describe an alternative picture to the Fregeanone—one which is equally an alternative to Russell or to TractarianWittgenstein—and to identify in it certain central principles which areincompatible with the Fregean view. That done, private language isthen introduced simply as an example of a violation of thoseprinciples. The point of the private language argument is to show thatit is the violation of those principles which makes private languageimpossible—which yields the result, that is, that private items couldnot be viewed correctly as having any semantics at all. The conclusionis that the principles in question are ones that we—or rathersemantics—could not live without. So they must be part of any correctpicture of semantics. Since they conflict with the Fregean picture, thatpicture must be wrong—at least to the extent that there is conflict.Since they require the main features of the Wittgensteinian alter-native, that alternative must be correct, at least with respect to thosefeatures. I now come to the most pleasant task of a preface—that of giving
xii Prefacecredit where it is due. Pleasant as it is, I think that there is more thanpolitesse involved here. For the credits reveal my own intellectualhistory, and hence the origins of the ideas expressed here. Seeing theirorigins may help to understand them better. To begin then, I wasgreatly inspired, and largely set in my present philosophical ways, bytwo of my undergraduate teachers at Berkeley, Thompson Clarke andHans Herzberger. Clarke gave me an appreciation of Wittgenstein andof Austin, and also gave me some form of the idea which appears inthis work as S-use sensitivity. Herzberger taught me a style ofapproaching problems which, while difficult to characterize, has beeninvaluable. His dispassionate way of setting out a problem issomething I can at least aim for, though probably not achieve. Inaddition, while still an undergraduate, I was helped along present pathsby Montgomery Furth at UCLA, and later, while a graduate studentthere, I was helped much further along them by Keith Gunderson. From the beginning, of course, I was greatly impressed by the worksof (the later) Wittgenstein and J. L. Austin. Unfortunately, I think,Austin has generally been read in quite the wrong way, so that in so faras he has been influential at all, it is quite peripheral aspects of hiswork which have captured the attention. That is a shame. There isnothing wrong with the idea of illocutionary forces, though not muchhas ever been done with it. But what matters most in Austin is hisviews on epistemology and on more central parts of semantics—viewswhich are the culmination of a tradition beginning with Cook Wilson,and moving (by way of reaction) through H. A. Prichard, and largelyechoed today in the writings of such philosophers as John McDowell.(So, one might say, an Oxford tradition despite itself.) More than justechoes of these views are to be found in the present work as well. Two other cases of action at a distance must be mentioned here.These are Hilary Putnam and Noam Chomsky. Again, both of thesemen have been influences on me from the very earliest days. I hopethat the influence of Putnam on this book will be obvious. From theearly sixties on, I have always thought that Putnam was one of the veryfew who understood what was important about Wittgenstein, and whofurther developed Wittgensteinian thought in proper directions. Hisphilosophical instincts have been—in my opinion—nearly infallible,and have moved his thought through the years in a natural progressionso that the fundamental insights of early articles such as WhatTheories Are Not5 remain intact and retain full value when placed inthe light of his current version of anti- (metaphysical) realism (aversion with which I find myself largely in agreement). Chomskys 5 Putnam [1962a].
Preface xiiiinfluence on the present work will be, I expect, less evident. I can onlysay that Chomsky presents a style of thinking about problems which Ifind not only immensely appealing but helpful. It is not so muchChomskys views which are in play here as—I hope, at least—his wayof organizing a topic. I now come to proximate causes. As noted at the outset, I would noteven have thought of writing this book, much less resolved to do so,were it not for John McDowell. His help did not end with that, but,applied regularly in small but highly concentrated doses, informed thework throughout. Next, I certainly would not have got the ideas hereto anything like their present degree of clarity or development—suchas that is—without the continued criticism, discussion, and some-times resistance put up by John Campbell. He has seen many drafts ofthis, and expended much effort on it. The result may not be what hecould unreservedly endorse,- but there would have been much less ofone without him. I have also been helped, in ways I would not want tohave missed, by the efforts, both critical and supportive, of HilaryPutnam and of Crispin Wright. Since the fall of 1986—thus, through-out the last three drafts—I have been further helped by discussionswith Michael Hallett, Jim Hankinson, and James McGilvray. Through-out, I have also benefited substantially from criticism by andnumerous discussions with Katherine J. Morris. This book is dedicated to my parents, Agnes and Philip Travis,whose invaluable boundless faith in me has never been conditional onrational justification for it—or on anything else.C. T.MontrealDecember 1987
Contents1 . Two Pictures of Semantics 12. The Making of Semantic Fact 373. The Uses of Language Games 824. Doubt and Knowledge Ascription 1295. The Limits of Doubt 1886. Through the Wilderness 2407. The Autonomy of Fact-stating 2938. The Problems with Private Semantics 337Bibliography 393Index 396
1 Two Pictures of SemanticsIn 243, Wittgenstein asks the following question:Would a language also be possible in which one could describe or express hisinner experiences—his feelings, moods, etc.—for his own use?—Then cantwe do that in our usual language?—But I dont mean it like that. The words ofthis language would have bearing on that which only the speaker could knowabout; on his immediate private sensations. Thus another person cannotunderstand this language.Thus he explicitly introduces the problem of private language.Whatever privacy is here, and whatever his arguments may be, onefixed point is that Wittgenstein holds private language to be imposs-ible. But why is that thesis interesting? This book can be seen as astory about why it is. It is a story about a small set of principles centralto Wittgensteins thought which are argued for by the way in whichprivate language turns out to be impossible. It is also and equally astory about a certain picture of semantics which those principlesrequire; a picture of the sorts of semantic facts there may be, ofsemantic properties, the ways in which items may have them, and theeffects and uses of their doing so. Chapters 2, 4, 5 and 6 will set out the principles involved, and showwhat philosophic work they do. They also aim to show that theprinciples are correct, and that it is reasonable to accept themindependent of any results of a private language discussion. But privatelanguage yields a different sort of argument in favour of them. Thestrategy, in broad brush strokes, is to test the principles by construct-ing a thought experiment in which they fail. If there are facts about thesemantics of private language, then what those facts are, and what itwould be to know them, cannot conform to the principles in question.The result, it will be argued, is that there can be no semantics for 1 Throughout, otherwise unidentified numbers refer to sections of WittgensteinsPhilosophical Investigations. The English translation sticks close to Anscombes, sincethat is the familiar one. But I have deviated from hers where I thought some philosophicpurpose required it, and where the German seemed as well or better served in thealternative way.
2 Two Pictures of Semanticsprivate language; that private language can express no judgements,much less ones which might be true (or false). Private language isimpossible because it is impossible for there to be either judgement orthought which is not governed by these principles. Otherwise put, thejoint assertion of semantic fact and denial of the principles for that factyields contradiction. The desired conclusion is that the principles arecorrect, and, specifically, that where items do have semantics, the factsabout their semantics, and the facts those items might express, aregoverned by those principles. Central to the principles is the denial ofwhat will here be called the dominant picture of semantics, as thatdenial is contained in what will here be called the alternative picture.Also central is the account, in the next chapter, of what makesjudgements about semantics correct. The desired conclusion is thatthe alternative picture, together with that account, apply whereverthere is semantic fact. Of course, on the description thus far, it is theprinciples as a whole which are tested by private language; conceivablythis particular part of them could have been idle in the collapse ofsemantic fact which takes place there. But as the story unfolds, we willsee that they are not idle. How these principles require the semantic picture to be developedhere, and how they clash with its alternative, will become clear whenthe principles are developed. This chapter aims at no more than sayingwhat the picture and its alternative are. The alternative has been thedominant picture at least since it was presented with great clarity,precision and elegance by Frege. Dominant, in fact, is an understate-ment. The picture is attractive enough to have been shared by nearlyeveryone, up to and including the present day. (Though Freges versionof a shared picture must be seen as containing proprietaiy elements.)The Wittgensteinian alternative to it, to be developed here, has had sofar little influence. This book aims to rectify that situation by showingwhat is intuitive and attractive about that alternative view, and, again,to do this independent of any effect the private language discussionmight have for it. But this chapter does not aim to do that. It aimsmerely to specify the defining characteristics of the picture, no matterhow unintuitive they may, for the moment, seem. The strategy will beto explain the dominant picture first, and then to develop the presentpicture by considering what one must do to disagree with that one.Disagreement, we will see, is no easy matter. Which may partlyexplain the attraction the dominant picture continues to exert. It bears emphasis that both pictures, being, as they are, pictures ofsemantic properties, concern whatever items may have them, and notexclusively words. They are at least as concerned with thought as withexpressions of it; with concepts as with predicates. The dominant view
Two Pictures of Semantics 3is just as likely to inform current discussion of belief and other suchattitudes, for example, as explicit discussions of language. And it islikely to inform such discussions by people who would not considerthemselves Fregean. On the other side, it is important always to keepin mind that the entire Wittgensteinian view of having semantics, as itwill be presented here and illustrated in the case of words, appliesintact to what words express and our attitudes towards that—so, interalia, to believing, meaning, intending, expecting and their objects.Further, semantic pictures have metaphysical correlates. The featuresthe present picture finds in predicates and the concepts they mayexpress are reflected in the properties we may have concepts of; thefeatures it finds in bearers of truth find their reflections in the factssuch items might state. By implication, at least, the dominant pictureis equally one of what properties and facts are like. Viewing the significance of private language as here viewed meansconstruing it as a sort of transcendental argument—a story about howthings must be, rather than merely about how they are—and one witha rather traditionally heavy metaphysical moral at that. It is hard toresist seeing the private language discussion as having some suchforce. But such argument does run counter to Wittgensteins statedviews on doing philosophy. To be impressed by how things must be, hetells us, is typically to be in the grip of a picture,- whereas what weought to ask is not how things must be, but how they are. So privatelanguage is an exceptional case. Part of the task in discussing it will beto explain why Wittgenstein is entitled to the exception. ]. SEMANTICSIt remains to be said what is meant here by semantics. The key notionis semantic property A semantics will be a set of semantic properties; asemantic item an item with semantic properties. It will not beimportant to give the notion semantic property sharp boundaries. Itwill be good enough if a semantic picture fits a large or important classof central cases. The notion semantic might then be broadened ornarrowed a bit to capture what the picture fits. But I will now list somecentral cases of semantic properties, and consider some principles forexpanding from these to new central cases. Some central semantic properties are ones of saying, or of havingsaid A (of B), and, less importantly, of meaning A, for any values of Aand B which yield sense here. Some other properties are saying A to beso, and saying A to be B. Still other properties are more specific oneswhich entail some of these, such as those of calling A (a) B, describing
4 Two Pictures of Semanticsor characterizing A as (a) B, or expressing the judgement that A is (a) B.A second class of central cases is epitomized by being true (false), beingtrue of A, saying what is true, and saying what is (would be) true of (insituation) A. Intuitively, true evaluates how an item fares in virtue of others ofits semantic properties—perhaps how, in virtue of them, it relates tothe world. So to be true, the thought goes, an item must have othersemantic properties of some appropriate sort. Whether words are true,for example, depends on what they say or said. So there must besomething which is what they do (did) say if they may be either true orfalse. Similarly, if a promise is kept, then whether it was depends onwhat was said would be done, so on what was said in the words thatmade it, so there must be something that was said in those words; if anorder was obeyed, then whether it was depends on what was said wasto be done, so there must be something said in the words which gave it;and so on. Call being true, being kept or obeyed, and the like evaluativeproperties. Then the general thought—intuitive, though perhaps notuniversally correct, is that if words or other items have evaluativeproperties, then they must have other non-evaluative semanticproperties; in the case of words, properties of roughly the firstmentioned sort. Strictly speaking, neither picture to be presenteddepends on this intuition being right. But it is a useful guide to furtherproperties one could recognize, which, if recognized, would besemantic. Further, there seems at least this much right about it: if anitem has an evaluative property—say, being true—then it is a sort ofitem which is liable to have some non-evaluative semantic propertiesas well; there are some such properties which it at least might have ormight have had. And if it instances an indefinitely large sort—thoughts, say, or words—then there are some such properties whichare had by some members of that sort, and members of the sortcharacteristically have some such properties. However, nothing oneither side of the issue to be presented requires assuming this. If itgrates on intuition, then let it be a purely optional assumption, usefulonly as an indication of some of what we will here recognize assemantic properties. From the above two core samples, the class of semantic propertiesmay be made to grow. Promises, orders and questions, for example, areneither true nor false. But orders and promises say things. An ordermay say that A is to do B. So orders and promises have semanticproperties of a general sort on which evaluative properties may depend.That is good enough reason (by fiat if necessary) to take them to havesome evaluative properties which depend on this semantics in the way
Two Pictures of Semantics 5that other evaluative properties such as being true depend on the samesort of semantics in other cases. Inter alia, orders may be obeyed.Whether a given situation counts as one of an order having beenobeyed depends on what the order says just as whether a givensituation makes words true depends on what they say. So we can takebeing obeyed (and similarly being carried out and being filled) asevaluative properties of orders. Questions do not say things. Neither dorequests. But requests may be carried out. That is an evaluativeproperty. So requests must have, as a rule, other semantic properties onwhich that evaluative property depends. Requests ask things. Theirbeing carried out by given doings depends on what they ask in the sameway as an orders being carried out depends on what it says. So askingA is a semantic property. But questions, too, ask things. So they mayhave evaluative properties in virtue of that. And so on. All this is nomore than intuition. But it is just these intuitions which pick out(roughly) the subject-matter of both the pictures to come. A semanticpicture need not fit all of these properties. We can revise intuitions ifnecessary to fit it. But the differences between present rivals will besettled in the centre of the field, and not on the periphery where suchrevision might be to the point. Let us now consider semantic items. The semantic propertiesconsidered so far have their most obvious applications in the case ofwords. Or almost. It is true that people can say things, and also meanthings, though perhaps not in a relevant sense of that treacherousword. On the other hand, people cannot be true (nor carried out) in therelevant sense. Are people semantic items? Say what you like aboutthat. That might be one reasonable way of describing their beliefs,intentions, expectations, and so on. It is a reasonable way of seeinghow Wittgensteinian principles apply to such things. But if it seemsunnatural to speak in this way, then legislate it away. There are more interesting things to say when we consider aproperty like being true. Words, perhaps, may have that property. Butthen so can (some of) what words express, or what they say, or what issaid in them. Such potential bearers of this property might be calledthoughts. As the name suggests, thoughts may be thought or believedas well as said or expressed. Or at least I will so use the word, since(some of) what we think and what we believe may be true. Sincethoughts may be true, they must have, as a rule, other semanticproperties on which their truth depends. Thoughts do not say things.What is said, for example, does not. So they must have other non-evaluative semantic properties which parallel in function thosesemantic properties of words. We will not yet worry about generalnames for those properties. But we note their existence. The important
6 Two Pictures of Semanticspoint is that the semantic pictures now at issue apply to theseproperties as well. Words may have the property of having been spoken at 3 p.m., andalso the property of having said that Montreal is an island. There is anintuitive difference between these properties: the second, but not thefirst, is a matter of how the words were to be understood. To knowwhen a given Mary had a little lamb was spoken, we need not knowwhich understanding of it was the proper one,- to know what it said, wemust know this. That P is a property which, for example, words havejust in case they are properly understood to have it might be thought tobe part of what characterizes P as semantic. That is a vague intuition ofas yet uncertain worth. In the next chapter, something specific will bemade of it within an account of which situations would be ones of anitems having one semantics or another. It will then turn out to pointto something characteristic of semantic fact. But it does not yet have aserious role to play. 2. THE DOMINANT PICTUREThe many accounts of semantics on the books differ in numerousrespects. The idea of a common picture is the idea of a common core inall, or virtually all, of these versions. The task of this section is todisplay that core in its most essential features. The strategy will beindirect. I will first consider a few intuitive, and, I hope, thoroughlyfamiliar ideas. These ideas contain a bit more than what is crucial forthe picture, though, I will suggest, very little more. But they point inthe right direction. Having presented them, I will then extract this corepicture from them. Though this book aims ultimately to treat all semantic itemsuniformly, it will help in the beginning to consider a plausibledistinction between items of two sorts. Intuitively, some items, likewords, or at least like English words (or German or Chinese ones, etc.),are liable to change their semantic properties. On one plausible view,there are other items which are not so liable: what was said in words,for example, or what was thought, or what is believed; in short, theitems we have labelled thoughts. Words may change their semantics atleast in so far as they may change what they mean over time. Ching isCantonese for green. But given that it comes to be used sufficientlydifferently by Cantonese speakers, it may come to mean blue. Onusual views, words may change their semantics in other ways too—over time, and perhaps across other parameters as well. On the otherhand, if, at a given time, Odile says, The walls are blue or The milk is
Two Pictures of Semantics 7sour, then what she said in those words cannot change its semanticsin any of these ways. And it is natural to conceive of it as not liable tochanging semantics at all, so as having whatever semantics it doeshave immutably. It is not as if what she said—for example, that themilk is sour—might be true today and false tomorrow. As we usuallyconceive things, either the milk is sour or it isnt. And that fact,whichever it is, settles the status of what she said once and for all.Similarly for any of its other semantic properties. It will help to treatthese two cases separately at first, and to begin with thoughts. The semantics a thought now has is, on the plausible view, asemantics it could not ever lack. But that is not yet to speak of thesemantics it could have had. In fact, the first intuitive thoughtcombines naturally with a second. Suppose I say that my chair is full offeathers, and the thought I thus express is true. For all that, it mighthave been false, and would have been, for example, if someone had letthe feathers out. The thoughts being true (or false) depends on how theworld is, and the world might have been different than it is in relevantrespects. On the other hand, it is natural to think that while some ofthe semantics of the thought might have been different than it is, otherparts of its semantic could not have been. The thought might havebeen false, given suitable co-operation from the chair, for example; butit could not have been a thought that the milk was sour, no matterwhat. Let us call those semantic properties that the thought could nothave lacked its intrinsic semantics. Here is the first intuitive idea to approach the core of the picture:some items, such as thoughts, have an intrinsic semantics; for suchitems, their intrinsic semantics fully determines their entire seman-tics. The intrinsic semantics of a thought fully determines thesemantics it now has, or, since it cannot change its semantics, thesemantics it has, full stop. Moreover, its intrinsic semantics fullydetermines the semantics the thought would have in any circum-stances which might have been the actual ones. Nothing other thanthe thoughts intrinsic semantics could be needed to determine, orcould play any role in determining, all the facts as to which semanticproperties that thought has in the actual situation which obtains apartfrom the thoughts semantics—the facts of matters other than itssemantic properties—no matter what that actual situation may be. Sofirst, if a thought, W, has a semantic property Q, then Ws intrinsicsemantics is incompatible with lacking Q, in this sense: there cannotbe an item with that intrinsic semantics about which one could state afact in saying W to lack Q; and, in particular, one could never state afact in saying W to lack Q. Such things might have been so, had theworld been different than it is,- but as it is, they cannot be so. Second,
8 Two Pictures of Semanticsfor any semantic property, Q, Ws intrinsic semantics cannot becompatible both with having and with lacking Q, in this sense: therecould not be both an item with Ws intrinsic semantics about whichone stated a fact in saying it to have Q and an item with that intrinsicsemantics about which one stated a fact in saying it to lack Q; and inparticular, it could not be both that one could state fact in saying W tohave Q, and either that one could state fact in saying W to lack Q, orthat one could ever fail to state fact in saying W to have Q. So there isno room for a further factor to decide which of these things is the case,given the world as it in fact is. If an intrinsic semantics does all thiswork, we will say that it is crucial for the semantics of the item whichhas it, and that that item has a crucial intrinsic semantics. Consider now words. For example, English words. Here the case iscomplicated by the fact that, as we have already decided, English wordshave no intrinsic semantics. At the very least, they might do suchthings as change meanings. But they might, in the actual world, varytheir semantics in other ways as well. Or it is plausible to think so. Onone view, English words, such as The chair is full of feathers, might,as it were, express different thoughts at different times. So that, if thechair was full of feathers yesterday but is empty now, one might holdthat those words were true yesterday but are false now (though thethought they expressed yesterday—that the chair was then full offeathers—and the different one they express today are both true at anytime). In that way, words might be seen as varying their semantics overtime. Words might also be seen as varying their semantics across otherand more general dimensions. Their semantics might vary acrossspeakings of them, for example—even across speakings produced atone time. As might happen with the English sentence I am hot, onone view of the matter. So that the semantics such words had asspoken in such-and-such speaking might differ from the semanticsthey had as spoken in another. All of these above sorts of variation are variations that can, perhapseven might, occur in the world as it actually is: the range of semanticswords would thus sometimes count as having are all semantics whichare consistent with the world being that way. But the last two sortsdiffer from the first. Intuitively they represent what words could doconsistent with their having the semantics they now do, and notmerely what they could do through ceasing to have that semantics.They might even be seen as representing what the words must do onother occasions given the semantics they have on this one; what thatpresent semantics requires of them elsewhere, given the changedvalues of the parameters across which their semantics varies. It isconsistent with the present semantics of The chair is full of feathers,
Two Pictures of Semantics 9for example, that it should express a different thought tomorrow thanit does today. In fact, this may even seem to be required by that presentsemantics. And, though it may have different semantics tomorrowthan it does today, those differences are only what its presentsemantics requires them to be. Similarly, the present semantics of Iam hot requires that it now express different thoughts in differentmouths. On all of those speakings it retains that present semantics,and acquires exactly that further semantics which this semanticsrequires. These ideas are somewhat rough. But, for present purposes,they are properly suggestive. Intuitively, some words sometimes express thoughts. When wordsdo so, their semantics is largely, if not entirely, that of the thoughtthey express (modulo some purely formal transformations.) Notably,the words are true (if we choose so to speak) iff the thought theyexpress is. Words, unlike thoughts, may have properties of expressingthought A, saying A, and so forth. But—trivially in the first case—which of these properties they have is determined, by formal rules, bythe thought they express, or by its semantics. Where words do notexpress a thought—as I am hot does not when not in a mouth, or as isfull of feathers does not, even in a mouth—they are plausibly seen asstanding to some item with an intrinsic semantics in the same waythat words which express thoughts stand to those items—the latterwords to a concept, perhaps, and the former to a meaning, orprepositional or thought schema, or something on these lines. All this suggests the following idea. The semantics of a thought (orother item with an intrinsic semantics is, on the first idea, determinedby its intrinsic semantics. Words cannot have an intrinsic semantics.But they could have a proper part of their semantics which does forthem what the crucial semantics of a thought is supposed to do for it.(We might think of this as a semantics which, if some item did have itintrinsically, would be a crucial intrinsic semantics for that item.) Wemight simply call this their crucial semantics. The idea is that they doso. Now, or on any occasion for words having a semantics—toutcourt, for example, or at a time, or on a speaking—those words have acrucial semantics which fully determines the semantics they have onthat occasion in the sense, described above, in which an intrinsicsemantics is meant to fully determine the semantics of an item (oritems) which have it. This is the first main intuitive idea about words. There is a second intuitive idea about words which is worthconsidering. It is suggested by this thought: the chair is full offeathers may express different thoughts on different occasions. Butsomething about the way it now is determines which thought it wouldexpress on any occasion where it did so—for example, at any time or
10 Two Pictures of Semanticson any speaking. Similarly for I am hot. The idea concerns the way inwhich this something does the determining it does. Take I am hot as amodel. Those words now have a certain semantics tout court. Theintuition is that, on a speaking, they retain essentially, though notnecessarily exactly, that semantics, and may acquire more. Forexample, they may then express a thought which is true (false)—something which they do not do tout court. The semantics they retainon a speaking may not quite be all of their semantics tout court. Onemight want to say, for example, that, tout court, they have the propertyof being neither true nor false. On a speaking, they may lose thatproperty, and gain instead the property of being true. The idea, in thiscase, is that the semantics they retain across occasions fully deter-mines their semantics on any such occasion, given the other facts ofthat occasion. The idea in general form is now this. Some words, now, or on someoccasion, have a semantics which they may retain across a variety ofoccasions for their having semantics. At least if they are unam-biguous—or on a reading of them if they are ambiguous—some part ofthis semantics, not necessarily a proper part, fully determines thesemantics they would have on any such occasion, given the other factsof that occasion; it is not compatible with their having, on thatoccasion, any of several divergent semantics (in the previously statedsense of compatibility). Since we have already appropriated the termcrucial, let us call this element of a semantics, for words that have it,a critical semantics. Like any semantics, a critical semantics issomething words are liable to lose: I am hot could, conceivably,change its meaning over time. The important point is what a criticalsemantics does for items that have it; what it determines as to whatwould be so of an item with that semantics. That words would saywhat was true if spoken in such and such way, for example, is onevariety of semantic property they may have. Their critical semantics iswhat fixes this variety of property for them. As conceived here, critical semantics differ from crucial ones. Therecould be items with a given critical semantics which differed in thesemantics they had; so in the semantic facts about them. It isconsistent with the world being as it is that such a variety of itemsshould actually be producible. There could not be items with a givencrucial semantics which did that. What there could not be, for a givencritical semantics, is an item, on an occasion for its having semantics,whose having that critical semantics was compatible with any ofvarious divergent sets of semantic facts about it on that occasion. Sowhile the critical semantics of I am hot is a semantics it has on aspeaking, it is not its crucial semantics on a speaking. But it fixes what
Two Pictures of Semantics 11that crucial semantics must be, given the other facts of that speaking. It is worth adding two notes to the above before moving forward.First, it is important to distinguish between the semantic determinacybrought about by crucial (or critical) semantics on the above thoughts,and what we might call the Fregean tertium non datui principle. Thislatter principle says that for any semantic item, W, and any semanticproperty, P, either W has P, or it lacks (does not have) P, tertium non datur. The ideas expressed above do not say that. For all that has beensaid so far, a given crucial semantics might fail to decide whether anitem has or lacks P. The important point is that if that is so, thennothing else decides the items status in this respect either. In thatcase, it is not a fact that the item has P, and we do not state a fact insaying so. But it is not a fact that the item lacks P, and we do not state afact in saying that either. In that case, the items having that intrinsicsemantics is incompatible both with its having P, and with its lackingP, in the stated sense. With Fregean tertium non datur, sets ofsemantic properties which might be the semantics of some item wouldalways have to be maximal sets: one could not add another semanticproperty to them without getting a contradictory set of properties. Butwe are not supposing that. Second, so far we have spoken of intrinsic or crucial semantics asdetermining the full semantics of an item. And we will continue so tospeak. But usually people have been interested in the determination ofa particular class of semantic properties, namely, evaluative ones. And,given the prominence of truth in philosophy, the properties specifi-cally at issue have usually been truth-involving ones—pre-eminentlyones of being true, of being false, of being true (false) of (item orsituation) A, and of being true (false) in (situation) A. Further, peoplehave often conceived of a crucial semantics as not containing suchtruth-involving properties. The second point is non-compulsory forpresent purposes. As for the first, no harm is done, for presentpurposes, if we restrict the picture in this way. Anyone who thinksthat that makes the dominant picture more likely to be right iswelcome to do so. Now for the core of the dominant picture. It seems to me that whatis most important in the ideas just sketched—and what will bepredominantly at issue in what follows—is not the idea that there aresuch things as crucial or critical semantics—that is, proper sub-parts ofa semantics which do (some or all of) the jobs just sketched—butrather the idea that there are such jobs to be done at all. The crucialidea, that is, is that a semantic item has a semantics which isdeterminate in the way just described. Here is the idea for thoughts (orany items whose semantics is immutable). Suppose that W is a
12 Two Pictures of Semanticsthought, and P any semantic property. Then we cannot work ourselvesin to each of a pair of situations, in one of which we would state a factin saying W to have P, and in the other of which we would state a factin saying W to lack P. Nor into a pair of situations in one of which wewould state a fact in saying W to have P, and in the other of which wewould not do so. Now consider words. Let W be any words, and O anyoccasion for their having a semantics. If words have a semantics at atime, then O may be a time. If they have a semantics on a speaking,then O may be a speaking of them. In general, O may correspond towhatever way in which it is correct to speak of W as having asemantics. (For example, in this book, "privacy" means . . .) Let P beany semantic property. Then we cannot work ourselves into a pair ofsituations in one of which we would state a fact in saying W to have(have had) P on O (at that time, on that speaking, etc.), and in the otherof which we would state a fact in saying W then to have lacked P, or atleast fail to state a fact in saying W then to have P. This, then, by current nomenclature, is the dominant picture. It is apicture of the way in which semantic items have whatever semanticproperties they have. (Again, without loss to the argument, it may berestricted to truth-involving properties.) We will find that it is a picturewhich Wittgenstein is centrally concerned to dispute. Let us say thatan item whose semantics is determinate in the above way has itssemantics classically, or is a classical item, or simply is classical. Thenon the dominant picture, all semantic items, or certainly all well-behaved ones, are classical. Wittgensteins view is that no item couldbe classical; to think that it could be is to make a mistake as to what itis to have a semantics. The rest of this book is about what kind ofmistake that is. Given this core, the idea of items having a crucial semantics can, Ithink, be added at no risk. Suppose, for example, that a thought has itssemantics classically. Suppose, now, that we posit a crucial intrinsicproperty of the thought—call it P. If we stick our necks out enough asto which property P is, then we might, perhaps, be refuted. But supposewe are willing to be open-minded as to what P is—that is, as to how itmay be identified other than as the crucial intrinsic property of thatthought. The only way in which we could be refuted, I think, is byfinding some other semantic property, Q—preferably a truth-involvingone—and showing that, if there is such a property as P, then it may bea fact that an item has P and has Q, and also may be a fact that an itemhas P and not a fact that it has Q. But that is just what we cannot dodecisively if the candidate thoughts semantics is determinate in thepresent sense. We cannot produce the required contrast with thatthought. And we cannot introduce some other thought into the
Two Pictures of Semantics 13discussion without running the risk that that thought lacks P. But,rather than insist on this idea, I simply drop the idea of crucialproperties from the core of the picture. On one natural view, semantic facts supervene on others. Considersome words, and an occasion on which they have semantics. All thenon-semantic facts of that occasion, or all the facts other thansemantic facts about those words—everything about how that oc-casion is, the semantics of those words aside—leave only one way forthose semantic facts to be. Or in so far as they do not—in so far as thereare, if any, then various sets of semantic facts compatible with all theother facts—there are no semantic facts about the words. The work tobe done here will challenge the idea, on at least one understanding ofit, that other facts leave only one way for semantic facts to be. But itwill equally challenge the idea that their not doing so provides anyreason to think there are no semantic facts. The world will exhibit, soto speak, an extra dimension along which semantic facts may vary; anextra factor at work in fixing them. We will then see that semanticfacts fail to supervene on others just in the way that any fact—inprinciple—fails to supervene on others. In a sense, there is nothingspecial about semantic facts, though they have a central role—here, atleast—in establishing the general point. On one view, there is a part of the semantics of a thought whichsupervenes vacuously on other facts: there is only one way for thatsemantics to be, full stop, since the thoughts having precisely thatsemantics is part of what it is for it to be the thought it is. In a sense,this idea, too, will be challenged. The point is not that a thought, forexample, that my chair is full of feathers, might have been anything atall. But the sort of indeterminacy we will discover in the semantics ofany item at all—thoughts included—is, inter alia, an indeterminacy inthe facts of what semantics a given thought could not have lackedwithout ceasing to be that thought; a variability in the propertieswhich identify it as being the thought it is. Which will mean that themost significant parts of what has usually been taken to be intrinsic toa thought are, in an important sense, not. And it will be shown thatthoughts cannot have a crucial intrinsic semantics—one which fullydetermines, in the present sense, what there semantics is. This willemerge as compatible with the idea that one can always view thesemantics of a thought as immutable, in the sense of not being subjectto change. We now have a view of what is so of what the dominant pictureapplies to. But we must note a qualification which may be built into iton what it is meant to apply to. It is reasonable to take it, at least, thatit is not meant to apply to inconsistent sets of semantic properties. So
14 Two Pictures of Semanticsalso, perhaps, not to items which cannot be assigned a consistentsemantics. Such items might count as vacuously fitting the descrip-tion of the picture provided here, since there would be no semanticproperty, Q, such that we could state fact in saying them to have Q.But they could not have such a thing as a crucial semantics. So theywould not fit those versions of the picture which posited one. Whichwould mean nothing. The picture was not made for them,- they mightas well be ruled outside the scope of the present discussion. Thinking along lines something like this, Frege, in at least some ofhis moods, and others following him, have held that the picture, or atleast their versions of it, is not meant to apply to everyday languagessuch as English or German, but rather to yet-to-be-constructedlanguages which would meet the requirements of serious and exactscientific discussion (however they would identify that). On that sortof move, Wittgenstein has the following comment (120):When I talk about language (words, sentences, etc.) I must speak the languageof every day. Is this language somehow too coarse and material for what wewant to say? Then how is another one to be constructed?—And how strangethat we should be able to do anything at all with the one we have!The comment reveals Wittgensteins aim. The point is not merely toshow that the dominant picture does not apply to this or that. It is notto show, for example, that ordinary language, though in order as it is,does not fit the picture. It is rather to show that there is no semanticitem which fits the picture, and, ultimately, that there could be none.That is certainly reason for wanting a transcendental argument, of thesort which I will claim the private language discussion to provide.(Though this does not yet explain how Wittgenstein is entitled to suchan argument.) Conversely, if the dominant picture fails for someparticular sort of item—English sentences, say—the defender of it stillwins if he can display some other sort of item—thoughts, say—forwhich it does not fail. This defines the ground rules for the ensuingdiscussion. 3. TWO FALSE STARTSIn arguing against the classical picture, one might adopt either of twogeneral strategies.2 One might try to show that the picture is senseless: 2 This sort of choice is often exhibited in the natural sciences. Suppose we takeseriously the idea that relativistic mechanics is the truth. Now consider the Newtoniandefinition of kinetic energy. We might say that it makes no sense, since it appeals tonotions of mass and velocity which make no sense. Or we might say that it is wrong,since, given what mass, velocity and energy must be, kinetic energy cannot be defined inthat way. (See Putnam [1962b] for discussion.)
Two Pictures of 5that somehow or other, in stating it, we have appealed to some(putative) notion that is not, and could not be well defined. Or onemight accept the picture as sensible and coherent, and argue that it isfalse—either of given semantic items, or, consonant with our am-bitious Wittgensteinian aims, of any semantic item. In this book I optfor the second more straightforward approach. But it must be said thatthe phenomena to be produced could be read as demonstrating the firstthesis. Again, this chapter will not argue that the dominant picture iswrong, but simply describe an alternative to it: a picture such that if itis right, then the dominant one is wrong. The main idea for an alternative picture will be a generalization ofWittgensteins intuition that, in some way, the present semantics of anitem leaves its future application—so at least some of its evaluativesemantic properties—open; that the application of a word is noteverywhere bounded by rules (84), the extension of the concept is notclosed by a frontier (68). This section will consider two developmentsof that thought which do not do the work of conflicting with thedominant picture, and one development which might do so, though wewill not pursue it here. The first development comes by viewing concepts, or their boundar-ies, as vague. If the concept is being a chair (or the words, is a chair),for example, then the point would be that there are some items whichneither determinately fit the concept (or those words), nor deter-minately fail to do so. It is just undecided whether they are chairs ornot. That observation, if correct, certainly counts against a Fregeantertium non datur principle. But it does not count against thedominant picture. For if the point is right, then nothing decideswhether the words or concepts are true of these items. So it is simplynot a fact that they have the property of being true of those items—onecould not state fact in saying them to have such a property—nor,equally, is it a fact that the words or concept are not true of the items.Grant all that, and the status of those words and concepts with respectto these truth-involving properties is perfectly determinate, in therelevant sense. So far, at least, they may be supposed to have theirsemantics classically. The first requirement for disagreeing with the picture, then, is thatthere be semantic facts—on the thought being developed, such thingsas facts as to a concepts future applications, or the future proper usesof words. It is then required, second, that some or all of these have anappropriate status with respect to the current semantics of the item(s)in question. (This underlines the point that one cannot disagree withthe picture simply by being a semantic sceptic, that is, by denying thatthere are semantic facts. As the present account will try to show,Wittgenstein was no such sceptic.) The appropriate status would seem
16 Two Pictures of Semanticsto require that, in some sense, whether these (candidate) semanticsfacts are facts must not be settled by the current semantics of the item.So, in an appropriate sense, it must not follow from that semantics thatthey are. On the other hand, if they are facts about the item, then thatthey are, it would seem, must follow in some sense from the itemscurrent status. So, it would seem, they—say, the relevant facts aboutfuture applications—are facts which, in a sense, do follow, and, in asense, do not follow from the items (current) understanding. AsWittgenstein points out,3 it is easy for one or both of these senses toappear to be a queer sense. Much of the current job is to explain themso that they do not appear queer. Let us try a second development of the thought. If the words is achair are true of an item, then they might have been false of it at leastin this way: with a sufficiently different history, chair might havemeant table-, the words might then have said what was false of theitem. Let us engage in a little (social) science fiction. Imagine that weare at the moment of the (fictional) introduction (by the EnglishAcademy) of the word chair into English. Not knowing what itsfuture history will be, we could not then say whether it would now saysomething true, for example, of my armchair. That fact might suggestan idea. We think that the meaning of chair is fixed somehow by itspast use. But why should we be biased in favour of that particular timedirection? Why should the semantics of chair not depend on the waysit will be used in the future as well? Future uses, for example, mightgive us reason to revise our views on what chair really said all along.4 Now let us apply this line of thought to the problem of semanticdeterminacy. Is is a chair true of my armchair? If that depends onfuture uses of it, then we might just have to wait and see. Or anyway,the future could, in principle, show us wrong if we answered, Yes. Soit is not determined yet, one might say, that it has that property. But itmight turn out to be determined some day that it does. That wouldthen be a semantic fact about those words which would not bedetermined by their current status. There may be much worth while in the core thought here. But again,it does not yet provide a way of disagreeing with the dominant picture.First let us distinguish two things that future use might be supposed todo. It might be supposed to determine that the words is a chair nowhave the property of being true of my armchair. Or it might besupposed to determine that at some time in the future they will havethat property. In the first case, we so far have every reason to say that 3 E.g., in 193, 194, 195 and 197. 4 Very sensible suggestions have been made in this direction by Crispin Wright—though not in this sort of case. See Wright [1986a], especially pp. 273-4.
Two Pictures of Semantics 17they have that property classically. Not all the events have yet takenplace which are required to make it so that they have the property.Still, that is what the course of the world will show. So, on thedescription so far, we would, in point of fact, state a fact in saying thosewords to have a property. And we would never fail to state a fact insaying this; nor would we ever state a fact in saying them to lack theproperty. Or so it seems. Consider the second case. By the samereasoning, the words will, in the future, have the property of being trueof my chair classically. As for the present, well, they just do not havethat property yet. So one would never state a fact in saying them nowto do so. Again, we have found no instance where the words do notnow, and will not in the future, have their semantics classically. There is another possibility. Suppose that what future uses deter-mine is neither of the things mentioned above, but rather this: that,while it is not now true to say that those words are now true of myarmchair, and one would not state fact in saying so, in the future it willbe true to say that those words now have the property of being true ofmy armchair, and one would then state fact in saying so. We await anaccount, of course, of how such a thing could be. But if it were so, as wewill see, it would be a good candidate for disagreeing with the picture.There would be something which in the future will count as a fact. Butthat something would not now (yet) count as a fact. The fact—when itdoes come to count as such—would be one about what the semanticsof the words now is, and what their now having that semanticsrequires. But one would not yet state truth in saying it to be a fact. There is the matter of the semantics of the words being what it is;and there is the matter of the semantics of the words being such-and-such. What can now be said truly on the latter score does not exhausteverything involved in the former fact. As one might put it, there is adistinction to be drawn between the fact of the items having itspresent semantics, and the fact that it has such-and-such semanticproperties. Such is what would be so on this line of thought. But herewe will not pursue it further in quite this form. 4. S-USEWittgenstein comments (432):Every sign by itself seems dead. What gives it life?—In use it is alive.Use, the suggestion is, is a factor on which the semantics of an itemdepends. Let us explore the idea that use may be an extra factor, inaddition to the semantics an item now has, on which the semantic
18 Two Pictures of Seman ticsfacts about it may depend; so a factor which may yield any of varioussets of such facts about the item, compatibly with its now having thesemantics it does. It is not easy to see how such a thing could be. But ifit could, that would explicate the sense in which current semanticsdoes not determine future applications. We will leave it to the nextchapter to explain the sense in which current semantics does do this.This chapter will consider the sense in which the future developmentis not already present (cf. 197); the next will explore the sense inwhich it is. The two taken together should remove the queernesswhich Wittgenstein says we may find in the idea that, for example, inunderstanding a word, we grasp its whole use. It is plausible enough that the semantics of some items depends ontheir use. I will ultimately argue that the semantics of all semanticitems does so. But use may refer to many phenomena. For themoment we want to consider one particular kind of use. In the corecase, it is the use made of words by a speaker in a particular speaking ofthem; for example, the use of the word banana made by Hugo insaying, The bananas are ripe. We will call such a use a speaker use,henceforth abbreviated to S-use. Presently, we will broaden the notionsomewhat. But for the meantime, let us confine attention to words.What we want to consider, then, is the relation of an S-use of words Wto the semantics of W, and to Ws semantics on that speaking. I want to call attention to a particular phenomenon—or, since I willnot yet argue for it,5 a possible phenomenon—which I will call S-usesensitivity. By itself, S-use sensitivity is not an alternative to thedominant picture. But it has a variety of important uses in what is tocome. Here is an example of it. Suppose that the refrigerator is devoidof milk except for a puddle of milk at the bottom of it. Now considertwo possible speakings, by Odile, of the words, Theres milk in therefrigerator. For the first, Hugo is seated at the breakfast table, readingthe paper, and from time to time looking dejectedly (but meaningfully)at his cup of black coffee, which he is idly stirring with a spoon. Odilevolunteers, Theres milk in the refrigerator. For the second, Hugo hasbeen given the task of cleaning the refrigerator. He has just changed 3 The problem is that for any putative example of S-use sensitivity, there are manynot obviously wrong strategies for avoiding the consequence that it really does exhibit S-use sensitivity. Nor are these strategies beyond the ingenuity of most philosophers. Theyreally need to be dealt with one by one. But doing that seriously would involve such along digression that we would never get on with the business of this chapter, nor of thisbook. (Though I will argue seriously for one instance of S-use sensitivity in chapter 4.) Ibegan to argue seriously for S-use sensitivity in Travis . I was initially impressedby Austins presentation of the phenomenon, primarily via his epistemological work,but also via his treatment of such examples as France is hexagonal in Austin . Mymost serious and systematic—and I hope successful—efforts at establishing thephenomenon are in Travis  and Travis .
Two Pictures of Semantics 19out of his house-cleaning garb, and is settling with satisfaction into hisarmchair, book and beverage in hand. Odile opens the refrigerator,looks in, closes it and sternly utters the above words. I claim that the example bears at least the following description:though there is no ambiguity in the English words There is milk in therefrigerator, or none relevant to the differences between the twospeakings, Odiles words in the first case said what was false, while inthe second case they said what was true. Both spoke of the same stateof the world, or the same refrigerator in the same condition. So, in thefirst case, the words said what is false of a refrigerator with but a milkpuddle,- in the second case they said what is true of such a refrigerator.Optionally we may also say that what was said in the words in the firstcase differs from what was said in those words in the second. Thespeakings may also differ with respect to yet other semantic properties.For all that, either speaking might have occurred, with that semantics,at any one given time—-10 a.m. on Sunday, say. I will not argue here that the above description of those speakings iscorrect. The present point, after all, is not to prove the dominantpicture wrong, but to identify something which would be at odds withit. Correspondingly, what we want to see at present is what wouldhappen if there were phenomena like the one described, either in thiscase, or in similar ones. In any event, where different speakings ofwords may thus differ in their semantics, consistent with their allhaving been made at some one given time, we will say that thesemantics of the words is S-use sensitive (with respect to the propertiesthat vary across the speakings). Since S-use sensitivity is so important for what follows, I willpresent one more initial illustration of it. Consider the words, Hugoweighs 79 kilos and the following situation: when Hugo steps on thescale in the morning, it reads 79 kilos, and that is a stable result.However, it is now after lunch; fully dressed (in winter clothing), Hugowould register 81 kilos on the scales. Now consider two speakings ofthe words. For the first, Hugo must weigh 79 kilos, and no more, toqualify for some sporting event. There is a discussion as to whether hedoes qualify. Odile, who has seen him step on the scale, tries to settlethe matter by revealing her information on the subject. She says, Hugoweighs 79 kilos. For the second, Hugo is about to step on to a verydelicate trestle bridge across a ravine which can take a maximum of 80kilos without snapping. Or Hugo is placed on a balance scale to weighout 79 kilos of gold (for some weighty purpose). The question iswhether Hugo ought to step on to the bridge, or whether that really is79 kilos of gold, and not more. Odile volunteers, Hugo weighs 79kilos. Again, the claim is that Odile spoke truth in the first case and
20 Two Pictures of Semanticsfalsehood in the second. So that the essential parts of the descriptionsof the first example fit this one as well. In that case, the words Hugoweighs 79 kilos also exhibit S-use sensitivity. (Similar examples couldbe constructed around a turkey which was said to weigh 5 kilos, anddoes so if you include the kilo of stuffing inside it. And so forth.) Herewe may say that the English words weighs 79 kilos, used as meaningwhat they do, and as speaking of weighing 79 kilos, may make any ofvarious distinct contributions to what is said in speaking them, withcorresponding differences in the conditions for the truth of what issaid. In this way at least, the words Hugo weighs 79 kilos havedifferent semantic properties on different occasions, and, as a result,used literally, may say both true things and false things of Hugo in agiven condition at a given time. That is, there are things of both sortsto be said in speaking them. Now I want to engage in a digression. Consider once again, I amhot. The intuition was that something about its semantics toutcourt—in particular, something about what it means—determineswhat its semantics would be on any speaking of it (where spoken asmeaning that). Now consider a different kind of case, as illustrated bythe ambiguous English sentence, Mary had a little lamb. Somespeakings of that sentence speak of meat eating. Some speak of petkeeping. Both sorts are consistent with what the sentence means assuch. We have the intuition, I think, that nothing about the meaning ofthat sentence—either tout court, or on a reading—determines itssemantics on a speaking in the way that the semantics of I am hotmight initially seem to determine its semantics on a speaking. (I willultimately argue that this initially plausible view is incorrect in thatcase as well.) It is not as if it were part of the meaning of Mary had alittle lamb that when certain other factors are present in a speaking itspeaks of meat eating, and when certain ones are present it speaks ofpet keeping. It is conceivable that some day someone should discoverrules to that effect. But nothing about our understanding of thesentence leads us to expect this. Nothing in that thought countsagainst the idea of critical semantics for words, for one thing, becausethat idea is explicitly restricted to unambiguous words. Though Maryhad a little lamb provides an idea on why it perhaps should be sorestricted. The point of the digression is this. I am hot and Mary had a littlelamb provide us with two paradigms of the variation of wordssemantics across speakings of them. Intuitively, S-use sensitivity, asillustrated so far, fits the paradigm of Mary had a little lamb and notthat of I am hot. On the present view of semantics, it does so. Nothingabout the meaning of Theres milk in the refrigerator, for example,
Two Pictures of 1identifies (in other terms) the conditions under which it would saywhat was true of a refrigerator with but a milk puddle. This thoughtdoes argue, I think, against the idea that words have a criticalsemantics. It certainly counts against the idea that one can account fortheir semantics on an occasion in terms of specifiable features orparameters of occasions in general, or indices, or something of the sort,and some function from values of these to at least some sort of crucialsemantics for words on a speaking, or a thought expressed in them (acontent in David Kaplans terminology6), where this function is fixedby the semantics of the words involved. However, rather thanfollowing out that line directly, I will let the idea of a criticalsemantics, like that of a crucial semantics, stand or fall for the samegeneral reasons that the idea of having a semantics classically willstand or fall. So, for example, aside from the specific arguments thatmight be brought against usual thoughts about indices and whatthese might do,7 there will be the general point that the semantics ofwords on a speaking is non-classical, hence there is indeterminacy inwhat the values of such things as functions over indices would have tobe in any given case. But this is to look ahead. S-use sensitivity must have metaphysical correlates. We will usethese correlates to broaden the notion. Suppose that, as is plausible, wetake Hugo weighs 79 kilos to speak of the property of weighing 79kilos, and to say Hugo to have it. Similarly, we might take Theresmilk in the refrigerator to speak of a property of containing milk, andto say the refrigerator to have it. For each sentence, we have considereda pair of speakings of it. If we identify what the sentence says (and sayson each speaking) in this way, then we must conclude the following.Hugo counts as having the property in question for the purposes of onespeaking, or for the purposes of evaluating it, and as lacking thatproperty for the purposes of (evaluating) the other speaking. Otherwise,given the description of what was said both times, there is noaccounting for the contrast in truth value. Similarly, the refrigeratorcounts as having the property in question for purposes of (evaluating)the one speaking, and as lacking it for purposes of (evaluating) theother. So the property of weighing 79 kilos is one Hugo sometimescounts as having (as he is at a given time, in a given condition), andsometimes counts as lacking (at that time.) Similarly for the refriger-ator and the property of containing milk. The sometimes here pointsto variation across occasions for considering whether Hugo (therefrigerator) has the property in question, or, if we like, acrossoccasions for using words to say him (the refrigerator) to do so. 6 See Kaplan . 7 For some of these, see Travis .
22 Two Pictures of Semantics When having a property, P, is liable to exhibit this kind of variationacross occasions, for some range of objects, we will say that having P isS-use sensitive, and that P is an S-use sensitive property. We might alsosay, if we choose, that, on the same reading of sometimes as above, itsometimes counts as a fact that Hugo weighs 79 kilos, and sometimesdoes not count as a fact, or counts as not a fact, that this is so.Equivalently, we might say that the fact that Hugo weighs 79 kilossometimes counts as obtaining and sometimes does not. When acandidate fact exhibits this feature, we may say that it is an S-usesensitive fact. Suppose we decided to say that there was an S-use insensitiveproperty of weighing 79 kilos, and that this is the property spoken of inthe words weighs 79 kilos. (If it is not, then it is difficult to see whatproperty those words would speak of, or what words might expressexactly that property.) Given the S-use sensitivity of the predicate, wewould then have to say that some speakings of it are strict and literal,so really do speak of that property, while others are not. For example,for the above pair of speakings by Odile, at most one member of thepair could exhibit a strict literal use of the words. It would then need tobe decidable, in principle, which member, if either, exhibited the strictliteral use, and decidable in general which speakings of the words didso. But first, the phenomenon of S-use sensitivity, if it exists, excludesthis possibility. It is built into the notion that if Odiles speakings doillustrate the phenomenon, then on both of them the words bear their(strict, literal) semantics. Second, it is apparent that there could be noprincipled way of deciding which of at least some contrasting pairs ofsuch speakings merited the privileged status. That is a point to beargued in arguing for S-use sensitivity. (An example of such anargument will be given in chapter 4.) But it is part of what is beingsupposed here in supposing S-use sensitivity (of words) to exist. One might respond to all this by saying that properties likeweighing 79 kilos, containing milk, being blue, etc., are actually notgood choices for properties. If they were properties, it might beallowed, they would be S-use sensitive ones. But that is enough reason,the claim would go, to suppose them not to be. We may say that thewords weighs 79 kilos speak of weighing 79 kilos. But strictlyspeaking, what they actually do is (roughly) to speak of a family ofproperties. A literal speaking of them, if fully intelligible and coherent,might speak of some one of these properties in particular. But thewords as such speak of all the members of this family indifferently. Itis worth considering why responses to the phenomena along theselines are at least highly likely to be bad ones. For the remainder of thischapter, I will then simply suppose them to be wrong.
Two Pictures of Semantics 23 If weighs 79 kilos or contains milk refers to a family of S-useinsensitive properties, the question is what the members of this familymight be. Suppose for the sake of argument that Odile spoke of someone of these properties in speaking truly of the refrigerator. Then wemay speak of the property Odile (then) attributed to the refrigerator.(Or we might prefer to speak of the property of the refrigerator invirtue of which what Odile said was true.) Call this property Q. Wenow have the means to attribute that property to other items, and toconsider whether other items, or the refrigerator in other states, haveit. Will Q be an S-use sensitive property? The key point is this. Indeciding that Odile spoke truly of the refrigerator, we solved oneproblem, or a few, about how to sort things into those containing milkand those not. But in principle there may always be more. In fact, wecan easily think of countless more: suppose that what was in therefrigerator was a slice of cheese or cheesecake, or a vial of secretions ofrabbit mammary glands, or a pint bottle of thoroughly coagulated (orvery sour) milk, or synthetic milk with the same molecular structureas milk but which had never seen a cow; and so on. (Similarly, considerthe property we decided Hugo has, even after lunch, if—but only if—his morning weighing showed—or suitable ones would show—79kilos. Would he have that property if the earths gravitational forcewere to be halved overnight? Or if, to take a Wittgensteinian example,he began occasionally to grow and shrink spectacularly and for noapparent reason?) Most of these problems, and others, are problems for Q as well. Andthe point is that it is reasonable (and so, I am claiming, correct) to solvethem in different ways on different occasions for doing so: their correctsolutions depend on the circumstances of the solving. Does arefrigerator with a slice of cheese or cheesecake on the bottom count ashaving Q? Yes, perhaps, if you are an orthodox Jew (engaged in certainactivities); no for many other purposes. (Note also that a vial of rabbitsmilk, whatever we think about its being milk, or an ordinary full pintbottle of milk in the refrigerator would not have been enough to makeOdiles true words true. Such things would not make the refrigeratorcount as having the right property for the purpose of assessing Hugoshouse-cleaning. There may be milk all right, but it is not in therefrigerator in the right sense; what it is in, of course, is vials andbottles.) On the evidence, Q will not be an S-use insensitive propertyeither. Nor will we make further progress towards expunging S-usesensitivity by repeating the move and speaking of some propertyattributed on occasion in speaking of Q. The thought was that if the property of containing milk would be S-use sensitive (if a property at all), then we may refine away the S-use
24 Two Pictures of Semanticssensitivity by speaking of some property of containing milk on someparticular understanding of what it would be to do that. The picturenow on offer, as illustrated in the above discussion, goes against thatthought. On it, we specify the property we want to speak of—theparticular understanding of being/doing such-and-such—however welike. There may then be, or arise, reasons both for and against takingsome item or items to have that very property we specified. It maythen be that on some occasions for considering those reasons, they arecorrectly (truly) judged to show the item(s) to have that very property,while on other occasions they would be correctly judged to show thesame item(s) to lack that property. No refining that we could ever docould guarantee that there could not ever be, in this way, two mutuallyconflicting and sometimes correct views of that very refinement andwhat it requires. I do not think we have any idea how to specify a property which, inprinciple, no item could have S-use sensitively. In any event, that ispart of the present view. First, we do not know how to specify aproperty such that problems could not arise as to whether an item hadit, given all the other facts about it, and all the reason they provide forcounting it as having the property, and for not doing so, where theseproblems did not have, tout court, uniquely most reasonable solutions.Second, we have no idea how to specify in advance what the solutionsto all such problems are to be in specifying which property it is.8Properties that we could not in principle specify cannot be the onesthat figure in our thoughts, words and judgements. If a given property, like the mentioned one of containing milk, is S-use sensitive, then its S-use sensitivity is unlikely to be eliminated byrestricting the range of items it applies to. The problems that may arisefor refrigerators, or others like them, would arise for breasts, bottlesand stomachs, or other potential containers of milk as well. Nor do weknow of some other item, rather like a refrigerator, and, as it were,cohabiting with it—a shadowy object, which we sometimes fill withmilk in putting milk in the refrigerator—which is not liable, inprinciple, to have the property of containing milk S-use sensitively.There might, perhaps, in fact be items about which there are not twothings to be said as to whether they contain milk. (Though these itemsare less easily found than one might think at first: the case of therefrigerator with the full pint bottle also contrasts with that of therefrigerator which is wall-to-wall milk, prepared to trap the unwary 8 Pioneering work in the direction of this point has been done by Hiitry Putnam andby f. L. Austin. Putnam does not go quite as far as the present point; Austin, I think, does.It is especially worth mentioning Putnam , a striking passage in Austin [1946, atp. 89 in the third edition], and the difficult but rewarding Austin [1952/3].
Two Pictures of Semantics 25opener of it in a deluge.) But these de facto univocally milk-containingitems, if any, are not ones which are not liable to be in a state in whichthey sometimes would, and sometimes would not, count as having theproperty in question. S-use sensitivity, the suggestion is, is an intrinsicfeature of the properties—and of the facts—in terms of which we thinkand speak about the world. It might reasonably be said to be put thereby something essential in the way we do that thinking. S-use sensitivity, as developed so far, does not force a break with thedominant picture. Nor is that the point of it. The picture might besaved, so far, in either or both of two ways. First, we might deny thatsuch things as English sentences—Hugo weighs 79 kilos or for thatmatter, Snow is white—have such properties as being true, at leasttout court. With Austin, we might deny that sentences were the rightsorts of things to be either true or false. If we like, we can continue tosay that they may have such properties as spoken in a given speakingof them. So far, it has not been shown that the semantics of words on aspeaking—the semantics of Odiles words in one of the above cases, forexample—is a semantics that such an item does not then haveclassically. Second, we might posit further semantic items, such aswhat was said in Odiles speaking of the words Theres milk in therefrigerator on some specific occasion, or the thought she thenexpressed. The item of such a sort which she thus produced on onesuch occasion need not be the one she produced on another. Again, wehave so far produced no reason for thinking that these items would notbe classical. It is important that, to follow this line, we must deny that an Englishsentence like Theres milk in the refrigerator may simply have theproperty of being true (of or in a given situation). For if a sentence mayhave or lack that property, then the fact that Odile might speak it truly(but literally) of a given situation shows that that sentence mustsometimes count as having the property, so that one sometimes statesfact in saying it to do so, while the fact that she might sometimesspeak it literally but falsely of the same situation shows the sentencesometimes to count as lacking the property of being true, so that onesometimes states fact in saying it to do that. In that case, the dominantpicture would not fit it, which at least gives us some idea what itwould be like for the dominant picture not to fit. Alternatively, we could just allow that the dominant picture doesnot fit English sentences. That would not be the end of the picture, socertainly not a victory for Wittgenstein. For there remain the otherposited entities: what was said on a speaking of a sentence, and so on.But we may now note something suspicious about this move. If arefrigerator is liable to have a property S-use sensitively, or if a property
26 Two Pictures of Semanticsis liable to be S-use sensitive as applied to refrigerators, or to that one,we are not likely to eliminate that possibility by finding further items,enough like refrigerators that they are eligible for having the propertyin question, but enough unlike refrigerators that they are not eligiblefor having it S-use sensitively. Could the matter really be fundamen-tally different with semantic items and their semantic properties?That is the question 1 will raise in the next section. To repeat,however, the existence of S-use sensitivity is not by itself inconsistentwith the dominant picture. Nor does the alternative picture consistmerely in positing it. The alternative picture has not been stated yet. Itis yet to come. 5. THE ALTERNATIVE PICTUREWe are now quite close to a description of how the dominant picturecould be wrong. There is but one more crucial step to take. Thissection describes that step. English is notorious for containing the means for its own semanticdescription. Some have thought that to be a bad thing. Good or bad,English contains the means for speaking of the semantic propertieswhich interest us, and which, rightly or wrongly, we sometimes takesome items to have. Any universal feature of English words, or theproperties they speak of, will be features of these semantic propertiesas well. Further, any such feature which is an essential part of speakingEnglish will also be a feature of any technical vocabulary by whichEnglish might be expanded. Notably, it will be a feature of any suchvocabulary specifically reserved for speaking of the property spoken ofin some English words on some given speaking of them—as illustratedby the account in the last section, if correct, of how the more refinedor precise properties we might hope to talk about in place of suchmundane ones as being blue or weighing 79 kilos, would still beinfected, in principle, with S-use sensitivity. So it would be a feature ofany semantic property we might speak of on occasion in speakingEnglish. If English has the right universal features, or even if enough ofthe right general features of English are essential features of itssemantic vocabulary and the properties that vocabulary refers to, thenit may be that the dominant picture is demonstrably wrong. Roughlyspeaking, if what goes for the property of weighing 79 kilos goes forsemantic properties such as being true, being true of my armchair, orbeing true of an item just in case it is red, then, we will see, thedominant picture is wrong. Using the last section as a source of clues,let us now consider what the right universal features might be.